DPRIETI Discussion Paper Series 20-E-069

How Do Voters Evaluate the Age of Politicians?

McCLEAN, CharlesUniversity of California, San Diego

ONO, YoshikuniRIETI

The Research Institute of Economy, Trade and Industryhttps://www.rieti.go.jp/en/

RIETI Discussion Paper Series 20-E-069

August 2020

How Do Voters Evaluate the Age of Politicians?1

McClean, Charles

University of California, San Diego

ONO, Yoshikuni

Tohoku University & Research Institute of Economy, Trade and Industry

Abstract

Elected officials tend to be older than most of the constituents they represent. Is this because voters generally prefer

older politicians over younger ones? We investigate this question by conducting two novel survey experiments in

Japan where we ask respondents to evaluate the photos of hypothetical candidates for mayor, and then alter candidate

faces using artificial neural networks to make them appear as if they are younger or older, while keeping their facial

structure and contours intact. Contrary to the observed candidate pool for mayors, the voters in our experiments

disliked elderly candidates the most, but viewed younger candidates as equally favorable as middle-aged candidates.

We also find that younger and middle-aged voters view candidates from their age group more favorably than others,

whereas older voters do not, and that all voters use age as a heuristic for a candidate's issue emphases and traits. We

then provide evidence for the external validity of our results using new data on actual mayoral elections. Together,

these findings suggest that it is supply-side factors rather than voter demand that explain the shortage of younger

politicians in public office.

Keywords: Elections, Voting Behavior, Public Opinion, Survey Experiment

JEL classification: D72, D91

The RIETI Discussion Papers Series aims at widely disseminating research results in the form of

professional papers, with the goal of stimulating lively discussion. The views expressed in the papers are

solely those of the author(s), and neither represent those of the organization(s) to which the author(s)

belong(s) nor the Research Institute of Economy, Trade and Industry.

1This study is conducted as a part of the Project “Research on Political Behavior and Decision Making: Searching

for evidence-based solutions to political challenges in the economy and industry” undertaken at the Research

Institute of Economy, Trade and Industry (RIETI). The author is grateful for helpful comments and suggestions by

Discussion Paper seminar participants at RIETI.

Candidates for political office vary significantly in age—some are young, some middle-

aged, some elderly—but in most countries, elected officials tend to be older than most of

the constituents they represent. Over half of the world’s eligible voters are under 40 years

old, compared to just 15% of national legislators (Inter-Parliamentary Union 2018). This

age bias is not limited to national institutions. In Japan, people under 40 comprise a third

of the voting age population but occupy less than 10% of the elected positions at any level

of government, from just 6% of municipal assembly members and 2% of mayors to 7% of

members of the House of Representatives.1

Recently, scholars have begun to pay more attention to the age bias of political insti-

tutions. One body of research discusses how the under-representation of younger people

in public office raises concerns about inter-generational equity (Bidadanure 2015; Gonzalez-

Ricoy and Gosseries 2016). Others provide evidence that just as the shortage of women or

racial and ethnic minorities in office can affect policy outcomes on issues related to gender

and race (e.g., Chattopadhyay and Duflo 2004; Franck and Rainer 2012), so too can the rel-

ative absence of younger politicians have consequences for age-related government spending

and social welfare programs (Alesina, Cassidy and Troiano 2019; McClean 2020; Curry and

Haydon 2018). Additionally, there is some evidence that gerontocracy can even discourage

younger people from participating in elections (Pomante and Schraufnagel 2015).

While studies of the causes of government by the elderly are relatively nascent, to date

most research on the topic has focused on supply-side explanations. In the U.S. context, two

recent books suggest that younger people have less ambition to become politicians because

they feel alienated from contemporary politics, view elected officials as corrupt, dishonest,

and inefficient, and believe that they can best enact change in their communities through

other means (Lawless and Fox 2015; Shames 2017). A handful of cross-national studies, on

the other hand, point to political institutions such as candidate-centered electoral systems

and minimum age requirements as the main culprits by restricting the ability of people to

1Calculated using the ages of politicians at the time of their election. Japanese House of RepresentativesElections Dataset (Smith and Reed 2018); Japanese Municipal Elections Dataset (McClean 2020).

1

run for office at a younger age (Joshi 2013; Stockemer and Sundstrom 2018).

In this article, we test a separate, demand-side explanation: that voters prefer older

politicians over younger representatives. We assess two channels though which age discrim-

ination could result in the under-representation of younger people in political institutions.

One possibility is that there exists a consensus among voters that younger candidates are

simply too inexperienced or not competent enough to hold public office. Another, not mu-

tually exclusive explanation is that voters prefer candidates closer to themselves in age, but

since younger citizens turn out to vote at lower rates than older people, the end result is

that younger candidates fare poorly in elections relative to older candidates.

In the second part of the paper, we consider two mechanisms that could account for

age-based discrimination in elections. The first is that voters infer information about a

candidate based on their age, from the policy issues the candidate will emphasize to the

traits they will exemplify in office, and that voters judge candidates based on these inferred

characteristics, whether they are right or wrong. In the economics literature, this perceived

link between a candidate’s ascriptive characteristics and their type is referred to as “statistical

discrimination” (Phelps 1972; Arrow 1973). The second mechanism that we explore is known

as “taste-based discrimination,” wherein voters may harbor negative prejudices about the

efficacy of certain groups as politicians, such as younger people, without any particular

rational explanation (Becker 1957).

To test for age biases and their mechanisms, we conduct two novel survey experiments

in Japan where we ask respondents to evaluate the photos of hypothetical candidates for

mayor, while at the same time we alter candidate faces using artificial intelligence to make

them appear as if they are younger or older. By taking advantage of recent advances in

machine learning and neural networks, we are able to hold constant elements of each photo

unrelated to aging—including the expression, underlying facial structure, and clothing of

candidates—and manipulate only those elements that tend to change with age. In the first

experiment, we randomly assign respondents to view two candidate photos and then ask

2

who they would vote for and whether they would turn out if these two candidates appeared

on the ballot. In the second, we present voters with individual candidate photos and ask

them to assess each candidate’s likely issue emphases, traits, physical attractiveness, and

electability.

Contrary to the observed under-representation of young people, we find that voters dislike

older candidates the most. Respondents were significantly less likely to say that they would

be willing to turn out or vote for older candidates, but were equally likely to support younger

and middle-aged candidates. This elderly bias was also reflected in our tests for in-group

favoritism: while younger and middle-aged voters were modestly more favorable toward

candidates from their age group compared to others, older voters were if anything more

critical of elderly candidates. We then test the external validity of our findings using newly

collected data on actual mayoral elections in Japan, and show that turnout is indeed lower

when candidates are much older and that older candidates tend to receive fewer votes than

younger candidates.

In our second experiment, we find support for both statistical and tasted-based mech-

anisms of age discrimination. Respondents in our surveys drew clear links between a can-

didate’s age and their issue emphases and traits. Voters believed that candidates would

devote relatively more attention to issues important to their age group: younger candidates

were thought to focus more on education and childcare, middle-aged candidates more on the

economy, and older candidates more on elderly care and healthcare. Middle-aged candidates

were seen as generally having the most favorable traits for office, whereas younger candidates

were viewed as the most physically attractive. In our subsequent tests, we find evidence that

inferred judgments about a candidate’s issues and traits are predictive of a voter’s support

for a given candidate (statistical discrimination), but that voters still significantly disliked

older candidates even when controlling for these other factors (taste-based discrimination).

In sum, we find little evidence that voter biases are to blame for youth under-representation

in office. Voters in our experiments were equally willing to support younger candidates as

3

middle-aged candidates, and preferred younger candidates substantially more than older

candidates. Our findings thus support the idea that it is supply-side factors such as politi-

cal ambition and institutions that most inhibit young people from becoming politicians. If

institutions can be reformed and young people encouraged to run, our results suggest that

voters will welcome their greater presence in public office.

Voter Biases and the Age of Politicians

To date, research on the informational shortcuts that voters rely on when they evaluate

politicians has largely ignored age. When evaluating a candidate, we know that voters make

inferences about their likely policies and effectiveness in office based on a wide range of

heuristics, from those more directly connected to politics, such as a candidate’s party and

endorsements (Rahn 1993; Sniderman, Brody and Tetlock 1991), to characteristics such as a

politician’s gender, race, class, religion, and even physical appearance (Aguilar, Cunow and

Desposato 2015; Burden, Ono and Yamada 2017; Calfano and Djupe 2009; Carnes and Lupu

2016; Todorov et al. 2005). However, despite significant media attention to the age bias of

political institutions,2 we know little about whether voters infer certain information based

on the age of a candidate and, if so, how it affects their decision-making at the ballot box.

Could the shortage of younger politicians be due to voters preferring older, more experi-

enced candidates over younger leaders? If voters generally dislike younger politicians, then

young people may choose not to run for elected office as often as older people, or they may

run but simply lose much more often at the polls.

Given the extensive research on voter biases concerning gender and race, it is easy to

imagine that voters might similarly discriminate against candidates based on their age.

Around the world, research shows that voters form stereotypes about female and racial

or ethnic minority candidates (Kam and Kinder 2012; McConnaughy et al. 2010; Kage,

2See for example Derek Thompson, “Why Do Such Elderly People Run America?” The Atlantic, March5, 2020, and Isabel Reynolds and Emi Urabe, “Japan Is Too Old-Fashioned, Says One of Youngest MinistersEver,” Bloomberg, September 11, 2019.

4

Rosenbluth and Tanaka 2019), and that these biases can help explain why so few women

and minorities hold office (Dolan 2004; Aguilar et al. 2015; Ono and Yamada 2020), although

they may play less of a role in contexts where other cues such as party affiliation are especially

influential (Aguilar, Cunow and Desposato 2015; Lawless and Fox 2015; Hopkins 2009).

If anything, voters may feel relatively more comfortable acting on age stereotypes because

in many contexts such biases are viewed as a more acceptable form of discrimination. For

example, while there are strong social norms that a person’s gender or race should not

exclude them from office, age is closely connected to experience, energy, health, and mental

acuity, all of which are seen as relevant grounds for judging a person’s ability to serve as

an elected representative. Moreover, while many countries have sought over time to remove

restrictions and encourage greater participation by women and racial or ethnic minorities in

elected office, codified discrimination against younger citizens has largely remained in the

form of legal rules that set the minimum ages for a person to be eligible to vote or stand

for office.3 These restrictions are similarly seen as more understandable because of concerns

about young people’s cognitive development, experience, and maturity, even though the

restrictions for certain positions can be as high as 40 years old for legislative offices and 50

for executive office in some countries.4

Age discrimination may be seen as more acceptable in part because age is often thought

of as a “different” type of social identity, one that is universally experienced and changes at a

constant rate with time. As Bidadanure (2015) writes, from a diachronic perspective young

people will not be mistreated over the course of their whole lives: even if age biases against

the young exist, everyone will be discriminated against equally when they are young, but

young people will eventually become older and have their time to serve in office. Others have

noted that age discrimination, while certainly an issue in many workplaces, does not come

with the same level of historical animosity, domination, and exclusion as other identities

3Some countries have recently lowered the age of candidacy to match the minimum age requirement forvoting, such as the United Kingdom, which lowered its age of candidacy from 21 to 18 in 2007.

4The minimum age of candidacy is 40 for the upper houses of Cameroon, Czech Republic, Rwanda, andZimbabwe, and 50 to be president of Italy.

5

such as race and gender (Phillips 1995; Mansbridge 1999).

In this paper, we focus on two sources of age bias that may contribute to the shortage

of younger politicians. The first is age stereotypes, by which we mean shared beliefs among

the electorate about the relative ability of individuals of different ages to serve in public

office. The second is in-group favoritism, wherein voters could prefer candidates closer to

themselves in age. The former could lead to a shortage if there is widespread dislike of

younger politicians among voters. The latter could lead to a shortage as a consequence of

older voters turning out to vote at higher rates than younger voters.

In terms of age stereotypes, there are good reasons to believe that voters may view some

candidates are “too young to run.” Criticisms of younger generations in the political arena

are common, in part because of their low voter turnout. As Holbein and Hillygus (2020,

7) write, young people have been described as “apathetic, disengaged, narcissistic, selfish,

entitled, shallow, lazy, impulsive, confused, lost, impatient, and pampered.” The conven-

tional wisdom among studies of low youth turnout is that young people lack the interest,

sense of civic obligation, or skills to participate in the electoral process, although they are

often quite active in other political activities (Wattenberg 2007; Dalton 2008; Holbein and

Hillygus 2020). It is possible that negative sentiments against younger voters could translate

to a negative bias against younger candidates. Even though younger candidates may be seen

as possessing some desirable qualities—from traits such as energy, technological expertise,

and longer time horizons to greater familiarity with issues important to young people such

as education, childcare, or climate change—we also know that voters value experience (e.g.,

Jacobson 1983; Shugart, Valdini and Suominen 2005). Younger candidates may therefore be

seen as simply too inexperienced and lacking the necessary skills to be viewed as competent

leaders when compared to older candidates.

There are also reasons to doubt whether voters truly prefer older politicians, however.

Outside of work in political science, there is an extensive literature on ageism in other settings

that finds age discrimination against the elderly is more common than biases against the

6

young (e.g., Kite et al. 2005). Studies on workplace discrimination have found that workers

share a widespread belief that at a certain point job performance begins to decline with

age (Kubeck et al. 1996; Gordon and Arvey 2004). While some view older workers as more

dependable and trustworthy, others see them as being more resistant to change, lacking

creativity, possessing fewer skills and less stamina, and being less willing to learn or work

with others (e.g., Posthuma and Campion 2009).

The combination of biases against the young and old could also result in a belief that

there is an “optimal” age for an elected official, a “Goldilocks” age that is neither too young,

nor too old, and strikes a balance between the benefits of youth and experience. Again,

in the workplace literature there are several examples where people perceive that there is

a “correct age” for certain positions. Retail, sales, technology, and finance are often seen

as particularly “young” industries (Broadridge 2001; McGoldrick and Arrowsmith 2001),

whereas jobs that require more managerial skills are typically associated with older workers

(Cleveland and Hollmann 1990). Negative stereotypes about workers are often strongest

when applicants do not match the perceived correct age for the position (Perry, Kulik and

Bourhis 1996). Thus, it could be that voters believe there is a best age for someone to serve

as their representative, and judge too young or too old candidates more harshly.

Alternatively, a second framework that could explain the under-representation of younger

people in office is in-group favoritism. People may feel the strongest connection to candidates

closer to themselves in age because people tend to identify more strongly with members of

their in-group, such as those who share a social identity (Tajfel and Turner 1986). As in the

literature on descriptive representation, there could be a belief that candidates who are closer

in age to a given voter will better emphasize policy issues that are important to members of

their age group (Webster and Pierce 2019). Since middle-aged and older people turn out to

vote at much higher rates than younger people, however, the distribution of politicians could

thus reflect the distribution of actual voters, rather than the distribution of the electorate.

As it stands, relatively few studies focus on voter attitudes toward the age of elected

7

officials, with two recent exceptions. The first is Pomante and Schraufnagel (2015), who run

an experiment with American college students where they randomly present respondents

with photos of candidates of different ages. The authors focus on turnout, rather than

vote choice, and find that younger people are more likely to say they will participate in

elections when there are younger candidates on the ballot. In the second, Webster and

Pierce (2019) use survey data from the Cooperative Congressional Election Study (CCES)

to gauge whether voters are more likely to support candidates closer to themselves in age.

The authors find some support for the use of age-based heuristics, although their use is most

prevalent in low-information elections, among higher educated individuals, and in judgments

made about co-partisan candidates.

These studies are not without their limitations, however. Both studies focus exclusively

on the United States, consider only in-group favoritism rather than general age stereotypes,

and confront some methodological challenges. Pomante and Schraufnagel (2015) rely on

student subjects, which means that we do not know how voters from other age groups would

react to their experiment. The authors also use photos of different candidates, who differ

in many ways apart from age, complicating their findings. Webster and Pierce (2019), by

comparison, rely on observational data, which has the added benefit of realism, but raises

concerns about selection bias. It might be that candidates from certain age groups share

other characteristics apart from age that make them more likely to receive support from

similarly aged individuals.

Although not focused on age, there is also some evidence concerning age biases from

studies using conjoint experiments that explore other research questions. Contrary to the

aforementioned studies, these experiments only test for overall bias, not in-group favoritism,

likely because age is generally included only as a control variable. The findings from these

studies are mixed, and sometimes conflicting. Depending on the setting and other variables

included, these studies alternately find that voters prefer younger candidates, middle-aged

candidates, or have no preference (Arnesen, Duell and Johannesson 2019; Clayton et al.

8

2019; Horiuchi, Smith and Yamamoto 2020; Kirkland and Coppock 2017; Kage, Rosenbluth

and Tanaka 2019; Ono and Burden 2019; Teele, Kalla and Rosenbluth 2018). In Japan, the

two conjoint experimental studies that include age have found that voters prefer younger

candidates for the House of Representatives, but paradoxically they also prefer candidates

with more experience such as incumbents that have held onto their seats for more than a

decade (Kage, Rosenbluth and Tanaka 2019; Horiuchi, Smith and Yamamoto 2020).

Conjoint experiments, too, have their strengths and weaknesses. Within candidate choice

experiments, conjoint analyses have a clear advantage compared to traditional vignette ex-

periments in their ability to control for a wide range of candidate characteristics and approx-

imate an information-rich environment (Hainmueller, Hopkins and Yamamoto 2014). One

disadvantage, however, is that the process of evaluating long lists of candidate attributes

against one another differs from the typical cognitive process that many voters go through

when evaluating candidates in real-world elections. Many voters do not collect such detailed

information on the set of candidates prior to voting, but instead rely on informational short-

cuts based on easy-to-observe characteristics of candidates such as their party, gender, race,

and, per our contention, age. Conjoint experiments can also present voters with unrealistic

profiles, such as younger candidates that have accumulated more terms in office than is likely

or even possible. Additionally, certain combinations may have interaction effects on voter

evaluations of candidate quality: younger candidates with lots of prior election wins may be

seen as superior when compared to older candidates that have failed to win even once.

We set out to test whether voters dislike younger or older politicians using an original ex-

perimental design in Japan. We see our experiment as a complement to conjoint analyses by

focusing on a comparably low-information environment that mimics the real-life process that

voters go through when they evaluate a candidate’s age via their appearance. Our focus on

age biases also enables us to test for both general stereotypes and in-group favoritism within

the same analysis, as well as the mechanisms underlying age discrimination in elections.

9

Evidence from Two Candidate Age Experiments in Japan

Japan provides an ideal setting in which to test for age biases against political candidates for

several reasons. First, young people have traditionally been under-represented in political

institutions across Japan, yet age-related policy issues are salient given that the country faces

the demographic challenges of a declining birthrate and rapidly aging population. Japan is

also a society where there are strong age norms regarding elder respect and the roles of

younger and older people in society.

Second, we focus in particular on mayoral elections, which involve candidate-centered

campaigns where nearly every candidate runs as an independent. Rather than relying on a

party label, candidates in these campaigns devote significant effort to developing a personal

vote. Campaigns in Japan are also famous for their strict regulations and heavy reliance on

campaign posters, which are posted throughout the municipality in high traffic areas and

feature above all else the candidate’s name and photo (McElwain 2008). Voters in Japan

are thus used to learning about candidates via these posters, which for many is their first

indication that an election is approaching (Lewis and Masshardt 2002). Furthermore, the

candidate pool for mayoral races is largely homogeneous—over 99% are ethnically Japanese

and over 98% are men—making age one of the most distinct differences between candidates.

Figure 1 shows the age distribution of mayors elected between 2004 and 2019 compared

to that of the voting age population. Japanese citizens must be at least 25 years old to run

for mayor, but the median age of an elected mayor is considerably higher at 62, 10 years

older than the median eligible voter (52). Figure 1 also reveals that younger voters are by

far the most under-represented age group: people under 50 make up nearly half (47%) of

the electorate, but just 9% of mayors. By contrast, middle-aged voters are clearly over-

represented: individuals between the ages of 50 and 70 comprise a third (32%) of eligible

voters, yet make up 79% of mayors. Lastly, elderly voters are also under-represented in office,

although to a lesser extent than younger voters. Citizens who are 70 or older represent a

10

Figure 1: Age Distribution of Mayors in Japan, 2004–2019

Eligible Age for Candidacy

0.00

0.02

0.04

0.06

20 30 40 50 60 70 80 90 100Age at Time of Election

Den

sity

Mayors Population

Sources: Japanese Municipal Elections Dataset (McClean 2020); Statistics Bureau of Japan (2004–2019).

fifth (21%) of the voting age population as opposed to 12% of mayors. The modal mayor in

Japan begins their term at 65 years old, the national age of retirement.

Finally, mayors in Japan have significant discretion over municipal policy. In evaluating

candidates, the advantage here is that survey respondents can expect that younger and older

candidates will have similar responsibilities in office regarding policy decisions. By contrast,

representatives in a legislature often have to work with other representatives, their party

leadership, and the executive in order to pass legislation. Moreover, given that legislatures in

Japan operate under strong seniority norms, respondents might expect that older legislators

(with potentially more experience in office) will have much more influence over policy than

younger legislators. Mayoral elections thus provide a setting where voters can believe that a

politician’s age has the clear potential to influence policy outcomes.

11

Experiment 1: Candidate Age and Voter Biases

To test for age biases, we fielded two experiments embedded in two different nationally

representative surveys in Japan. Experiment 1 was administered in March 2020 by Rakuten

Insight Inc., one of the major survey companies in Japan. We randomly selected our survey

sample from Rakuten’s subject pool after adjusting their settings to match the population

census in terms of respondent age, sex, and region of residence. We initially aimed at

collecting a sample size of 3,000 but received more than 3,000 valid responses.

For Experiment 1, we began by purchasing licenses from Shutterstock to use and manip-

ulate the photos of two different male Japanese models (Table 1). We selected these models

because they looked similar to typical mayoral candidates and their photos looked similar

to typical campaign materials such as candidate posters. In the photos, both models are

wearing dark grey suits with brightly colored ties, have slight smiles, face directly toward the

camera, and have relatively conservative haircuts that could be seen on people of different

ages. One of the models is additionally raising a clenched fist in a sign that is commonly

used by candidates of all ages in Japanese elections.

After purchasing these photos, we next manipulated them to appear younger or older

using FaceApp. FaceApp is a free, mobile application created by Wireless Lab for both iOS

and Android that became popular in summer 2019 because of its ability to realistically age

or de-age a user’s photos. While such applications had existed before, FaceApp became es-

pecially popular thanks to its photorealism and its use by many celebrities. On social media,

the app went viral on Twitter and Instagam under the #AgeChallenge, which challenged

people to upload images of themselves with the app’s old-age filter applied. By mid-July,

more than than 150 million people had downloaded the app.5

While FaceApp uses a proprietary algorithm that Wireless Lab does not share publicly,

5The application also came under controversy for privacy concerns about how the Russian company wasusing an individual’s photos. See for example John Koetsier, “Viral App FaceApp Now Owns Access toMore Than 150 Million People’s Faces and Names,” Forbes, July 17, 2019.

12

Table 1: Hypothetical Candidates for Mayor

Younger Middle-Aged Older

Model 1

Model 2

Notes: In Experiment 1, respondents were randomly shown two candidate photos at the same time, one ofeach model. In Experiment 2, respondents were randomly shown one candidate photo.

the company has said that it ages or de-ages photos using artificial intelligence and neural

networks. More specifically, this process involves manipulating faces along several dimen-

sions, including changes to (i) wrinkles, especially on the forehead, above the nose, and in

the smile lines between the nose and mouth; (ii) skin elasticity, as skin becomes looser as

we age, especially underneath the eyelids and around the neck; (iii) color contrast, as faces

with high color contrast between the eyes, lip, and mouth tend to appear younger than faces

with low contrast; (iv) skin pigmentation, as hormones and sun exposure darken the skin

over time; and finally (v) hair color, as our hair follicles tend to grow grey, silver, or white

as we age.

Importantly, FaceApp only manipulates elements of each photo that are likely to change

with aging, but does not modify the model’s underlying facial structure, expression, hair,

or anything related to their clothes or the background of the original photo. Moreover,

the manipulated photographs retain sufficiently high resolution such that each photo looks

realistic. Using such photos in our experiment thus effectively controls for any potential

13

confounding factors unrelated to aging that might arise from using different models or actual

candidate photos. This is especially important given the extensive literature on how aspects

of a candidate’s appearance such as attractiveness, smile, facial structure, and skin tone can

influence voter evaluations (Atkinson, Enos and Hill 2009; Bailenson et al. 2008; Ballew II and

Todorov 2007; Banducci et al. 2008; Berggren, Jordahl and Poutvaara 2010; Caruso, Mead

and Balcetis 2009; Horiuchi, Komatsu and Nakaya 2012; King and Leigh 2009; Kirkland

and Coppock 2017; Krupnikov, Piston and Bauer 2015; Lawson et al. 2010; Terkildsen 1993;

Todorov et al. 2005; Weaver 2011)

We used FaceApp to create three versions of each model’s photo that approximated the

age range of mayoral candidates in Japan, which are shown in Table 1. For each model,

we aimed to create one photo on the younger end of the spectrum, one middle-aged, and

one older. In a pilot survey (sample size: 300), we asked respondents to guess each model’s

age to verify that the photos met our expectations. Respondents estimated that the three

versions of Model 1 were 37, 62, and 82 years old, and that the three versions of Model 2

were 33, 58, and 80 years old. We also checked respondent estimates of each candidate’s

age in the survey itself (sample size: 3,000), and ended up with slightly younger estimates:

respondents thought that the photos of Model 1 were 29, 52, and 72 years old, respectively,

and that the three photos of Model 2 were 30, 51, and 72 years old.

For Experiment 1, we randomly assigned respondents to view two photos (one of each

model) from Table 1 and told respondents that that they would be asked a series of questions

about a mayoral election in which these two people were the candidates. We additionally

mentioned that (i) neither candidate was the incumbent; (ii) both candidates were indepen-

dents; and (iii) the election was for a city where the respondent resides. Our experiment

is thus a 3×3 factorial design, where the three levels of each treatment are the younger,

middle-aged, and older versions of each model’s photo.

After showing the two candidate photos, we then asked each respondent two questions:

whether the respondent would turn out to vote in an election featuring these two candidates

14

on the ballot, and which candidate the respondent would vote for in the election. Our two

main goals were to test whether voters are more likely to turn out or support candidates of

a certain age group compared to others (age stereotypes) and whether they view candidates

closer to themselves in age more favorably (in-group favoritism).

Our research design does have some disadvantages relative to existing candidate choice

experiments, such as those rely on vignette or conjoint analyses. The first is that we cannot

control for other relevant characteristics of candidates that might affect voter evaluations or

interact with age, such as a candidate’s work history or stated issue positions. The second

is that we lack some specificity with regard to the candidate’s age. Whereas vignette and

conjoint analyses can list the exact age of candidates, we ask respondents to make their own

judgements based on photos.

Yet, we believe that our design offers several advantages and serves as an important com-

plement to other experimental work. The first is the added realism of relying on candidate

faces rather than researcher-created stories about the candidate or long lists of candidate

attributes in paired tables. We seek to approximate how voters might react when they see

a candidate in person or in their campaign materials. Our experiment thus emulates a sen-

sation that is closer to a first impression or a low-information environment where voters do

not necessarily know a lot about each candidate’s background or issue positions. Our study

is closest to that of Pomante and Schraufnagel (2015), but we improve on their research

design by using manipulated photos of the same person rather than different models and

photos that differ along multiple dimensions apart from age. The innovation to our design is

that we use advances in machine learning to create realistic photos that would not otherwise

be possible, allowing us to capture voter evaluations about candidates at different stages in

their life cycle within an experimental setting.

15

Figure 2: Candidate Ages and Turnout

●

●

●

●

●Older vs. Older

Middle−Aged vs. Older

Younger vs. Older

Younger vs. Middle−Aged

Younger vs. Younger

(Baseline: Middle−Aged vs. Middle−Aged)

−30 −20 −10 0 10Percentage Point Change in Probability of Voter Turning Out

Notes: Dependent variable is equal to 1 if a respondent said they would turn out for an election with thesecandidates on the ballot, and 0 otherwise. Baseline turnout is 79%. Bars show 95% confidence intervals.

Do Voters Dislike Younger or Older Politicians?

Do voters hold certain biases toward candidates based on their age? To find out, we begin

by considering how our candidate photos affected respondent willingness to turn out in a

hypothetical election. For ease of presentation, we average our results across the two models

such that Younger Model 1 vs. Middle-Aged Model 2 is treated the same as Middle-Aged

Model 1 vs. Younger Model 2. We thus collapse our 3×3 experiment into six treatment

conditions: Younger vs. Younger, Younger vs. Middle-Aged, Younger vs. Older, Middle-

Aged vs. Middle-Aged, Middle-Aged vs. Older, and Older vs. Older.

Figure 2 plots the likelihood of respondents saying that they would turn out to vote

across these treatment groups, using turnout in Middle-Aged vs. Middle-Aged elections

(79%) as the baseline condition. The first four treatments are strikingly similar and do not

differ significantly from the baseline group. So long as at least one candidate was younger or

middle-aged, voters responded that they would be just as likely to participate in the election.

By comparison, respondents were 15 percentage points less likely to report that they would

16

Figure 3: Candidate Age and Vote Choice

●

●Older

Younger

(Baseline: Middle−Aged)

−60 −40 −20 0 20Percentage Point Change in Probability of Voter Choosing Candidate

Notes: Dependent variable is equal to 1 if the respondent said they would vote for the candidate, and 0otherwise. Bars show 95% confidence intervals.

turn out when both candidates were older. In sharp contrast to the idea that voters prefer

older politicians over younger ones, we find that voters are significantly less interested in

mayoral contests with only older candidates.

Does this bias against elderly politicians affect vote choice as well? Figure 3 plots the

difference in the probability that respondents would vote for the younger or older candidate,

using the middle-aged candidate as the comparison condition. We again average our results

across both models, but this time focus only on choices made by respondents in treatment

conditions where the candidates in the hypothetical election differed in age: Younger vs.

Middle-Aged, Younger vs. Older, and Middle-Aged vs. Older.

As shown in Figure 3, we find that voters disliked older candidates substantially more

than either younger or middle-aged candidates. When presented with a mayoral race between

an older candidate and either a younger or middle-aged candidate, voters were more than

50 percentage points less likely to choose the older candidate. In contrast, voters appeared

equally willing to cast their vote for either the younger or middle-aged candidate.

Taken together, these findings thus suggest that voters have a strong, negative bias

against elderly politicians, but do not dislike younger politicians compared to other age

groups. While demand-side factors may play a role in explaining the steep drop-off in the

number of candidates that run for office after age 70, our results suggest that voter biases are

not to blame for the shortage of younger candidates. Voters do not dislike younger candidates

17

Figure 4: Candidate Ages and Turnout by Age of Respondent

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

Younger Voters Middle−Aged Voters Older Voters

−30 −20 −10 0 10 −30 −20 −10 0 10 −30 −20 −10 0 10

Older vs. Older

Middle−Aged vs. Older

Younger vs. Older

Younger vs. Middle−Aged

Younger vs. Younger

(Baseline: Middle−Aged vs. Middle−Aged)

Percentage Point Change in Probability of Voter Turning Out

Notes: Dependent variable is equal to 1 if a respondent said they would turn out for an election with thesecandidates on the ballot, and 0 otherwise. Baseline turnout is 73% for younger voters (under 50), 83% formiddle-aged voters (50–69), and 89% for older voters (70 and over). Bars show 95% confidence intervals.

any more than middle-aged candidates, even though mayors under 50 are outnumbered nearly

9 to 1 by mayors between the ages of 50 and 70 (Figure 1).

In-Group Favoritism

A second pathway through which age biases could help explain the shortage of younger

candidates is in-group favoritism. If voters view candidates closer to themselves in age more

favorably, but older voters turn out at much higher rates than younger voters, then the

end result could still be an electorate-level bias toward older candidates. To explore this

possibility, we break down our experimental results in Figure 4 into three groups depending

on the age of the respondent: younger voters (under 50), middle-aged voters (50–69) and

older voters (70 and over).

We again find that the first four treatment conditions, where at least one candidate is

younger or middle-aged, are not significantly different from the baseline group (Middle-Aged

vs. Middle-Aged) for any respondent age group. Where we do see some difference across

18

Figure 5: Candidate Age and Vote Choice by Age of Respondent

●

●

●

●

●

●

Younger Voters Middle−Aged Voters Older Voters

−60 −40 −20 0 20 −60 −40 −20 0 20 −60 −40 −20 0 20

Older

Younger

(Baseline: Middle−Aged)

Percentage Point Change in Probability of Voter Choosing Candidate

Notes: Dependent variable is equal to 1 if the respondent said they would vote for the candidate, and 0otherwise. Bars represent 95% confidence intervals.

age groups, however, is with the Older vs. Older treatment: younger voters experience the

largest, most significant drop in turnout compared to middle-aged or older voters. Younger

voters, who already have the lowest baseline turnout—both in our survey and in actual

elections—are also the ones most turned off to voting by an election that features only

elderly candidates.

In Figure 5, we similarly break down our results for vote choice by age of respondent. We

find some evidence of in-group favoritism for younger and middle-aged voters, but not for

elderly voters. When comparing younger and middle-aged candidates, we find that younger

voters are 8.6 percentage points more likely to choose the younger candidate (p<0.01),

whereas middle-aged voters are 5.6 percentage points more likely to vote for the middle-aged

candidate (p=0.06). By comparison, the right panel in Figure 5 reveals that older voters do

not feel significantly more attached to older candidates than other age groups, and in fact

tend to dislike them even more than younger voters.

Given that middle-aged voters turn out at higher rates than younger voters, could this

level of in-group favoritism help explain the shortage of younger candidates? In the Ap-

pendix, we re-estimate the vote choice analyses shown in Figure 3 for likely voters, i.e., those

who self-reported that they would turn out for the race (Figure A1). While likely voters

chose middle-aged candidates more often than younger candidates, the estimate is small (1

percentage point) and not statistically significant. In comparison, the negative bias against

19

elderly candidates is even larger at more than 60 percentage points.

In sum, our findings again suggest that voter biases are unlikely to play a substantial

role in explaining the under-representation of younger politicians in mayoral offices, but they

may help explain the relative absence of much older politicians.

Evidence from Actual Mayoral Elections

Experiments tend to have high internal validity, but they can suffer from low external validity.

While we believe that our experiment approximates the real experience of Japanese voters,

there are of course clear differences between viewing a candidate photo via an online survey

and seeing a candidate in person or on a poster on the street, not least because the latter

person is an actual candidate for office.

In this paper, we test whether our findings generalize to actual elections using the

Japanese Municipal Elections Dataset (JMED) assembled by McClean (2020). This newly

created dataset includes information on the near universe of candidates who competed in

mayoral elections between 2004 and 2019, including information on their age, gender, in-

cumbency, and vote share as well as turnout in the election. Using this data, we focus on

competitive races and set up both election- and candidate-level analyses where the dependent

variables are turnout and vote share, respectively. To avoid the methodological complication

that candidate vote shares in the same election are not independent of one another, we focus

on the top-two candidates in races between incumbents and challengers and then estimate

each regression analysis separately. Our key independent variables are the sum of the two

candidate’s ages for our turnout analysis and each candidate’s age for the vote share regres-

sions. Finally, for each analysis, we first estimate the bivariate relationship and then a model

that includes controls for gender, incumbency, and municipality and year fixed effects.

The results in Table 2 provide further evidence in support of our survey experiments.

In terms of turnout, we find in our fully specified Model 2 that turnout declines as the

sum of candidate ages increases. Our within-municipality estimate suggests that for every

20

Table 2: Candidate Age, Turnout, and Vote Choice in Actual Mayoral Elections

DV: Turnout DV: Vote Share

Elections Incumbents Challengers

(1) (2) (3) (4) (5) (6)

Sum of Candidate Ages 0.183∗∗∗ −0.051∗∗∗

(0.027) (0.017)Age of Candidate −0.319∗∗∗ −0.555∗∗∗ −0.131∗∗∗ −0.211∗∗∗

(0.035) (0.053) (0.027) (0.037)Female −2.409∗∗∗ −2.864 −6.611∗∗∗

(0.585) (3.450) (1.354)Incumbent −1.586∗∗∗

(0.322)Constant 41.685∗∗∗ 79.733∗∗∗ 48.572∗∗∗

(3.279) (2.222) (1.588)

Observations 3,388 3,388 2,879 2,879 2,710 2,710R2 0.018 0.930 0.036 0.636 0.009 0.644Municipality Fixed Effects No Yes No Yes No YesYear Fixed Effects No Yes No Yes No Yes

Notes: Analysis is for the top-two candidates in competitive races for mayor (2004–2019). Standard errorsclustered by municipality are shown in parentheses. ∗p<.1; ∗∗p<.05; ∗∗∗p<.01.

additional 10 years of age among the top-two candidates, turnout declines by approximately

0.5 percentage points. Interestingly, the sign of this effect is reversed in our model without

controls or fixed effects. This implies that there is a relationship between the types of

municipalities or election years that tend to have higher turnout and the average age of

candidates that contest mayoral elections. At the municipality level, this finding makes

sense given that turnout is often very high in smaller towns and villages in Japan (Horiuchi

2005), where candidates for mayor also tend to be older (McClean 2020).

Turning to vote share, we find that older candidates for mayor tend to receive a sig-

nificantly lower percentage of the top-two candidate vote share. For every 10 years of a

candidate’s age, our results suggest that incumbents receive as much as 5.6 percentage

points less in vote share (Model 4), whereas the vote share of challengers declines by 2.1

percentage points (Model 6). These are substantively large effects, and hold even when con-

trolling for other characteristics of candidates and time-invariant factors specific to certain

21

municipalities and election years.

Although not the main focus of this study, we also find in Table 2 that turnout is sig-

nificantly depressed by 2.4 percentage points when there are female candidates in the race

(Model 2). Moreover, while female incumbents tend to receive the same number of votes as

their male counterparts (Model 4), female challengers receive 6.6 percentage points less on

average (Model 6). These results deserve further attention in a separate study, but in this

paper they provide an interesting context with which to compare age and gender biases. We

find that on average, the voter bias in turnout against candidates that are 10 years older is

about one fifth that of the bias against female candidates (0.5 vs. 2.4 percentage points). As

for vote choice, age biases appear to be more influential than gender biases when it comes to

incumbents, as the latter is not statistically significant, whereas the bias against challengers

that are 10 years older is roughly a third that of the bias against female challengers (2.1 vs.

6.6 percentage points).6

Experiment 2: Mechanisms of Age Discrimination

Why do voters dislike elderly candidates? Why do younger and middle-aged voters like

candidates closer to themselves in age, but older voters do not? As discussed earlier, we

focus on two potential mechanisms of age discrimination: (i) statistical discrimination, where

voters use age as a heuristic with which to infer information about the expected performance

of an older politician in office, whether accurate or not, and then evaluate the politician based

on this information; and (ii) taste-based discrimination, where voters just plainly dislike older

candidates without any clear explanation.

To test for the influence of these two mechanisms, we fielded a second experiment with

a nationally representative survey in Japan. Experiment 2 was administered in March 2020

as part of the “Survey on Attitudes Toward Politics, Society, and the Economy,” which was

conducted by the Research Institute of Economy, Trade, and Industry, a policy think tank

6Here, the observed gender bias in actual elections runs counter to the null findings found in conjointexperiments in Japan (Kage, Rosenbluth and Tanaka 2019; Horiuchi, Smith and Yamamoto 2020)

22

in Japan. Our sample size was again 3,000 respondents.

In Experiment 2, we consider three pathways through which this age discrimination

against elderly candidates may operate. The first is that voters may infer that older politi-

cians will focus on issues less important to many younger or middle-aged voters, or perhaps

that they have less energy to focus on issues in general. The second is that voters may infer

that older people have less favorable traits and will be less competent as politicians. The

third, which does not necessarily relate to statistical discrimination, is that older candidates

may be disliked because they are viewed as less physically attractive (e.g., Todorov et al.

2005). We then compare to what extent these factors predict a voter’s belief about a candi-

date’s electability (statistical discrimination) and test whether there is any residual dislike

of older politicians after controlling for these factors (taste-based discrimination).

More specifically, in the experiment we randomly assign respondents to view a single

candidate photo and then ask them questions about the candidate’s likely issue emphases,

traits, attractiveness, and electability. We mentioned that the candidate was running for

a mayoral election for a city with a population around 300,000. We chose to specify the

population of the city as municipalities in Japan range significantly in size, from a population

less than 100 to more than 3 million. Mayors of cities with populations above 200,000 are

granted additional discretionary powers by the central government over municipal policy.

We decided to ask about a relatively larger city to avoid having respondents make their own

inferences about the type of municipality and level of mayoral discretion. Each respondent

completes the experiment twice, once for each model, with each experiment thus being a

1×3 design.

Policy Issues

When voters see candidates of different ages, do they infer that these candidates will em-

phasize different policy issues? One possibility is that voters associate politicians more with

issues that are especially salient for members of the politician’s age group. Thus, voters may

23

believe that younger politicians will focus more on education and childcare; middle-aged

politicians may be seen as especially concerned with the economy; and older politicians may

be linked most to elderly care and healthcare. If these issues are important to voters, then

voters may prefer candidates in their age group because they believe these candidates will

devote the most attention to these age-related issues. A second possibility is that voters as-

sociate age with overall energy level and attention to policy in general. Voters may therefore

dislike elderly candidates more than middle-aged or younger ones because they believe that

older candidates will devote the least energy to policy issues.

To assess these hypotheses, we asked respondents how likely they thought each candidate

would be to emphasize 11 different policy issues. We selected these issues based on our own

substantive knowledge about local government in Japan and by consulting past elite and

public opinion surveys. For each issue, respondents answered on a 5-point Likert scale,

ranging from “Very Unlikely” to “Very Likely” to emphasize the issue. In the main text, we

focus on the percent of respondents who said that the candidate would be “Likely” or “Very

Likely” to emphasize the issue.

Figure 6 plots the average responses for our 11 policy issues together with a 12th plot

that calculates the average level of issue attention by candidate age group. As with our

earlier experiments, we average the results across our two models.7

We find clear age differences across these policy issues. Respondents in general thought

that younger candidates would be more likely to focus on education, childcare, environment

and climate change, anti-corruption, and foreign residents and multiculturalism. Middle-

aged candidates were most associated with the economy and unemployment, public works,

budget deficit, and crime and safety, Finally, older candidates were seen as emphasizing

elderly care and healthcare.

We can also see that the most popular issues that voters connected with candidates

relate to welfare and the economy. It is notable that the top two issues receiving focus

7We find no substantial differences if we instead calculate the results separately for each model.

24

Figure 6: Candidate Age and Policy Issues

Crime and Safety Elderly Care Healthcare Average (All Issues)

Foreign Residents Economy and Employment Public Works Budget Deficit

Education Childcare Climate Change Anti−Corruption

0

10

20

30

40

50

60

0

10

20

30

40

50

60

0

10

20

30

40

50

60

Mea

n S

ayin

g C

andi

date

Wou

ld B

e Li

kely

to E

mph

asiz

e th

e P

olic

y Is

sue

(%)

Younger Candidate Middle−Aged Candidate Older Candidate

Notes: Dependent variable is equal to 1 if respondents said that the candidate would be “Likely” or “VeryLikely” to emphasize the policy issue, and 0 otherwise. Bars represent 95% confidence intervals.

across candidates were those that are most important to elderly voters: elderly care (49.3%)

and healthcare (45.7%). Education (41.9%) and childcare (38.8%) did follow close behind,

but our results still indicate that on average voters expect politicians to pay more attention

to elderly issues. We also see the biggest differences in voter inferences on these age-related

welfare issues: voters saw younger candidates as more than 20 percentage points more likely

25

to focus on education and childcare compared to older candidates, and likewise more than 20

percentage points less likely to emphasize either elderly care or healthcare relative to older

candidates.

Issues related to the economy came in second to welfare concerns, such as the economy

and employment (37.1%), public works (33.1%) and budget deficit (27.4%). For these issues,

however, we find a curvilinear relationship where voters associated the middle-aged candidate

most with economic issues. This finding could reflect voter perceptions that middle-aged

candidates strike a balance between having more experience with the economy compared to

younger candidates, yet still being active participants in the labor force compared to older

candidates.

Among the issues receiving somewhat less attention, we can also see that voters clearly

connected younger candidates with more liberal issues such as climate change (29.7%), anti-

corruption (21.2%), and foreign residents and multiculturalism (19.0%). The age gap for

climate change is especially large, with voters perceiving younger candidates as being more

than 15 percentage points more likely to emphasize the issue relative to older candidates.

Finally, we find in our last plot that older candidates are seen as devoting the least

attention to issues overall by about 5 percentage points relative to the other age groups.

While we conduct more formal tests later in the paper, this perceived lack of attention to

policy issues—especially those apart from elderly care and healthcare—could be part of the

story for why older candidates are viewed so negatively by voters.

Traits

Do voters believe that older politicians lack the traits necessary to be effective in office?

Figure 7 explores the link between a candidate’s age and respondent evaluations about their

likely traits. We asked about eight different traits in all, and again include a final plot that

shows the overall average across traits.

We find that middle-aged candidates are generally viewed the most favorably, with some

26

Figure 7: Candidate Age and Traits

Experienced Long−Term Oriented Average (All Traits)

Considerate Competent Consensus−Oriented

Dominant Reliable Determined

0

10

20

30

40

0

10

20

30

40

0

10

20

30

40

Mea

n S

ayin

g C

andi

date

Fits

this

Tra

it (%

)

Younger Candidate Middle−Aged Candidate Older Candidate

Notes: Dependent variable is equal to 1 if respondents said that the trait was “Applicable” or “Very Appli-cable” to the candidate, and 0 otherwise. Bars represent 95% confidence intervals.

exceptions. Voters inferred that middle-age candidates would be more likely to be reliable,

determined, competent, and consensus-oriented than other age groups. Younger candidates

were seen as the most considerate, albeit by a small, insignificant margin, and the least

dominant. Older candidates were seen as the most experienced and the least long-term

oriented.

27

Figure 8: Candidate Age and Attractiveness

Attractiveness

0

10

20

30

40

50

Mea

n S

ayin

g C

andi

date

is A

ttrac

tive

(%)

Younger Candidate Middle−Aged Candidate Older Candidate

Notes: Dependent variable is equal to 1 if respondents said that the candidate’s appearance was “VeryGood” or “Good,” and 0 otherwise. Bars show 95% confidence intervals.

Perhaps unsurprisingly, we find the largest difference in experience. Respondents were

over 35 percentage points less likely to report that the younger candidate should be seen

as experienced compared to the older candidate, and over 25 percentage points less likely

compared to a middle-aged candidate.

Thus, while the average result shown in the last box suggests that younger candidates

are evaluated least favorably with regard to traits, this effect is largely driven by experi-

ence. Compared to older candidates, for example, younger candidates are seen as signif-

icantly more determined, competent, consensus-oriented, and long-term oriented, whereas

older candidates are seen as more dominant.

Attractiveness

Are older candidates viewed as less attractive than other candidates? Figure 8 plots the

results from our question asking survey respondents whether they found their randomly

assigned candidate photo to be attractive. As suspected, younger candidates are seen as the

most attractive, followed closely by middle-aged candidates, with older candidates lagging

28

behind by more than 25 percentage points.

Electability

The previous three sections have found that voters draw clear links between a candidate’s

age and their issue emphases, traits, and attractiveness. Across all three categories, older

candidates tended to fare less favorably, with the exception of questions about elderly issues

and experience. Do these three factors explain why voters tend to dislike older candidates?

Or do voters still dislike older candidates even when controlling for these characteristics?

We put these questions to the test by examining the individual-level correlation between

respondent answers to these three questions, the respondent’s own age, their estimate of

the candidate’s age, and a final question where they rated the overall electability of the

candidate. As with previous survey questions, the electability question was asked using a

5-point Likert scale, but here we dichotomize the responses such that the dependent variable

equals to 1 if the respondent said the candidate was “Likely” or “Very Likely” to be elected

mayor, and 0 otherwise. Table 3 reports the results from separate OLS regression analyses

for each candidate age, averaged across the two models.

Across these three models, we can see that voter inferences about a candidate’s policy

issues, traits, and attractiveness are significant predictors of their opinions about the candi-

date’s likely success, with the exception of policy issues for older candidates. We find that

each additional policy issue (up to 11) raises voter expectations about a candidate’s electabil-

ity by 1.7 to 2.1 percentage points for younger and middle-aged candidates, respectively, but

has no significant effect on older candidate evaluations. The impact of traits is considerably

larger, and has the biggest effect on younger candidates, with each associated trait (up to

8) raising their electability rating by 7.7 percentage points. Finally, attractiveness is by far

the strongest predictor, increasing evaluations by 19.0 to 31.3 percentage points depending

on the candidate’s age.

While these results offer support for our statistical discrimination mechanism, the find-

29

Table 3: Determinants of Candidate Electability by Age

DV: Electability of Candidate

Younger Candidate Middle-Aged Candidate Older Candidate

(1) (2) (3)

Estimate of a Candidate’s Age −0.001 −0.004∗∗∗ −0.005∗∗∗

(0.001) (0.001) (0.001)Policy Issues 0.017∗∗∗ 0.021∗∗∗ 0.005

(0.004) (0.004) (0.003)Traits 0.077∗∗∗ 0.067∗∗∗ 0.046∗∗∗

(0.007) (0.006) (0.006)Attractive 0.255∗∗∗ 0.313∗∗∗ 0.190∗∗∗

(0.021) (0.024) (0.025)Age of Respondent 0.0002 0.0003 −0.001∗∗∗

(0.001) (0.001) (0.0005)Constant 0.017 0.212∗∗∗ 0.374∗∗∗

(0.037) (0.073) (0.078)

Observations 2,053 1,944 2,003R2 0.346 0.429 0.237

Notes: Dependent variable is equal to 1 if respondent said the candidate was likely to be elected mayor, and0 otherwise. Standard errors clustered by respondent are shown in parentheses. ∗p<.1; ∗∗p<.05; ∗∗∗p<.01.

ings for estimates of a candidate’s age also suggest an effect of taste-based discrimination.

Even when controlling for voter inferences about policy issues and traits, as well as voter

perceptions of the candidate’s attractiveness, middle-aged candidates and older candidates

that were seen as older were significantly more disliked compared to those seen as younger.

For older candidates, their electability rating dropped on average by 5 percentage points

for every 10 years in their age estimate. By contrast, voters did not show any evidence

of taste-based discrimination against younger candidates. Those who viewed the youngest

models as significantly younger did not view them more negatively, and the point estimate

in fact suggests they may have been viewed more favorably.

Lastly, we also find in Model 3 that older respondents were significantly less likely to view

older candidates as electable. This matches with our lack of a finding for in-group favoritism

in Experiment 1. In other words, not only are older candidates viewed the least favorably,

but they tend to be viewed less favorably by older voters.

30

Together, these results provide some evidence for both statistical and taste-based dis-

crimination. Evaluations based on age about a candidate’s issue emphases and traits were

positively correlated with whether that respondent saw the candidate as electable. However,

for middle-aged and older candidate photos, the respondent’s estimate of the candidate’s

age was still negative and strongly significant even when controlling for these other factors,

suggesting a residual dislike of elderly candidates.

Discussion

Young people are significantly under-represented in most political institutions. Our find-

ings in this paper, however, indicate that most voters would be happy to see this age bias

corrected. Across our experimental and observational tests, we find that voters are equally

supportive of younger candidates as middle-aged candidates, and actually prefer younger

candidates significantly more than older candidates.

Our study thus has implications for policymakers interested in expanding younger peo-

ple’s presence in political institutions. We find no evidence that demand-side explanations

such as voter biases pose significant hurdles to more young people serving in elected office.

While more studies are needed to replicate our results and test the robustness of supply-side

explanations, the extant evidence suggests that age biases in institutions can best be ad-

justed through tackling younger people’s lack of political ambition and reforming institutions

to make it easier for them to run for office (Shames 2017; Lawless and Fox 2015; Stockemer

and Sundstrom 2018; Joshi 2013).

Apart from overall age stereotypes, we also find evidence that voters employ both sta-

tistical and taste-based age discrimination in elections. Voters infer clear policy emphases

and traits based on a candidate’s age, which in turn are linked to their perceptions of the

candidate’s electability. Younger candidates may not be viewed as having as much experi-

ence as other age groups, but they are seen as the most likely to emphasize a wide range of

issues in office, from education and childcare to climate change, anti-corruption measures,

31

and policies that benefit foreign residents and promote multiculturalism. By contrast, we

find a sharp drop-off in voter support for candidates as they become elderly: voters described

older candidates as less competent than others despite their additional years of experience

and saw these candidates as the least likely to focus on any policy issue outside of those

most important to elderly voters.

Our results thus add new dimensions to the nascent, but growing literature on age biases.

While Pomante and Schraufnagel (2015) find that younger Americans are more likely to turn

out when younger candidates are on the ballot, we find in Japan that this is true for all voters

when compared to races with only elderly candidates. Younger people do not turn out more

when younger people are on the ballot compared to middle-aged candidates, which represent

more common mayoral races, but they do turn out significantly less when mayoral candidates

are older. As for Webster and Pierce (2019), we find similar evidence of age-group favoritism

in our experimental results as they did in their observational study. One caveat of our

findings is that this in-group favoritism does not hold for elderly voters, who tend to be

more critical than others of elderly candidates.

Our findings also contribute to studies of elections in Japan that have touched on the age

of candidates, albeit tangentially. While Horiuchi, Smith and Yamamoto (2020) and Kage,

Rosenbluth and Tanaka (2019) find that younger candidates are generally the most liked

in races for the House of Representatives, we find that voter preferences are more balanced

between younger and middle-aged candidates for mayoral races. We also find much stronger

negative biases against older candidates. These differences in our results could reflect our

differences in method, but they may also be indicative of the fact that we focus on mayors

who are local executives with much greater discretion over policy compared to legislators.

Future research is needed to disentangle how age biases might differ across government levels

(local vs. national), positions (legislative vs. executive), and levels of policy influence.

Finally, we hope that our new experimental design can inspire additional work on age in

other countries and electoral contexts. While conjoint analyses have exploded in popularity

32

in recent years, no single research design can be a panacea for answering every research

question. Moreover, our design only scratches the surface with regard to what is possible in

terms of image manipulation via machine learning and neural networks. As these processes

grow increasingly sophisticated and accessible to scholars, more research is needed to explore

how people react when confronted with realistic-looking representatives (or other political

actors) of different ages, as well as how age biases interact with discrimination against other

identities, such as gender and race.

33

References

Aguilar, Rosario, Saul Cunow and Scott Desposato. 2015. “Choice Sets, Gender, and Can-

didate Choice in Brazil.” Electoral Studies 39:230–242.

Aguilar, Rosario, Saul Cunow, Scott W. Desposato and Leonardo Barone. 2015. “Ballot

Structure, Candidate Race, and Vote Choice in Brazil.” Latin American Research Review

50(3):175–202.

Alesina, Alberto, Traviss Cassidy and Ugo Troiano. 2019. “Old and Young Politicians.”

Economica 86:689–727.

Arnesen, Sveinung, Dominik Duell and Mikael Poul Johannesson. 2019. “Do Citizens Make

Inferences From Political Candidate Characteristics When Aiming for Substantive Repre-

sentation?” Electoral Studies 57:46–60.

Arrow, Kenneth J. 1973. The Theory of Discrimination. In Discrimination in Labor Markets,

ed. Orley Ashenfelter and Albert Rees. Princeton University Press pp. 3–33.

Atkinson, Matthew D., Ryan D. Enos and Seth J. Hill. 2009. “Candidate Faces and Election

Outcomes: Is the Face-Vote Correlation Caused by Candidate Selection?” Quarterly

Journal of Political Science 4(3):229–249.

Bailenson, Jeremy N., Shanto Iyengar, Nick Yee and Nathan A. Collins. 2008. “Facial

Similarity Between Voters and Candidates Causes Influence.” Public Opinion Quarterly

72(5):935–961.

Ballew II, Charles C. and Alexander Todorov. 2007. “Predicting Political Elections From

Rapid and Unreflective Face Judgments.” Proceedings of the National Academy of Sciences

104(46):17948–17953.

Banducci, Susan A., Jeffrey A. Karp, Michael Thrasher and Colin Rallings. 2008. “Ballot

Photographs as Cues in Low-Information Elections.” Political Psychology 29(6):903–917.

34

Becker, Gary S. 1957. The Economics of Discrimination. University of Chicago Press.

Berggren, Niclas, Henrik Jordahl and Panu Poutvaara. 2010. “The Looks of a Winner:

Beauty and Electoral Success.” Journal of Public Economics 94(1-2):8–15.

Bidadanure, Juliana. 2015. “Better Procedures for Fairer Outcomes: Youth Quotas in Par-

liaments.” Intergenerational Justice Review 2(15):40–46.

Broadridge, Adelina. 2001. Ageism in Retailing: Myth or Reality? In Ageism in Work and

Employment, ed. Ian Glover and Mohamed Branine. Burlington, VT: Ashgate pp. 153–174.

Burden, Barry C., Yoshikuni Ono and Masahiro Yamada. 2017. “Reassessing Public Support

for a Female President.” The Journal of Politics 79(3):1073–1078.

Calfano, Brian Robert and Paul A. Djupe. 2009. “God Talk: Religious Cues and Electoral

Support.” Political Research Quarterly 62(2):329–339.

Carnes, Nicholas and Noam Lupu. 2016. “Do Voters Dislike Working-Class Candidates?

Voter Biases and the Descriptive Underrepresentation of the Working Class.” American

Political Science Review 110(4):832–844.

Caruso, Eugene M., Nicole L. Mead and Emily Balcetis. 2009. “Political Partisanship Influ-

ences Perception of Biracial Candidates’ Skin Tone.” Proceedings of the National Academy

of Sciences 106(48):20168–20173.

Chattopadhyay, Raghabendra and Esther Duflo. 2004. “Women as Policy Makers: Evidence

from a Randomized Policy Experiment in India.” Econometrica 72(5):1409–1443.

Clayton, Amanda, Amanda Lea Robinson, Martha C. Johnson and Ragnhild Muriaas. 2019.

“(How) Do Voters Discriminate Against Women Candidates? Experimental and Qualita-

tive Evidence From Malawi.” Comparative Political Studies 53(3-4):601–630.

35

Cleveland, Jeanette N. and Genevieve Hollmann. 1990. “The Effects of the Age-Type of Tasks

and Incumbent Age Composition on Job Perceptions.” Journal of Vocational Behavior

36(2):181–194.

Curry, James M. and Matthew R. Haydon. 2018. “Lawmaker Age, Issue Salience, and Senior

Representation in Congress.” American Politics Research 46(4):567–595.

Dalton, Russell J. 2008. The Good Citizen: How A Younger Generation Is Reshaping Amer-

ican Politics. Washington, D.C.: CQ Press.

Dolan, Kathleen. 2004. Voting for Women: How the Public Evaluates Women Candidates.

Boulder, CO: Westview Press.

Franck, Raphael and Ilia Rainer. 2012. “Does the Leader’s Ethnicity Matter? Ethnic Fa-

voritism, Education, and Health in Sub-Saharan Africa.” American Political Science Re-

view 106(2):294–325.

Gonzalez-Ricoy, Inigo and Axel Gosseries, eds. 2016. Institutions for Future Generations.

Oxford: Oxford University Press.

Gordon, Randall A. and Richard D. Arvey. 2004. “Age Bias in Laboratory and Field Settings:

A Meta-Analytic Investigation1.” Journal of Applied Social Psychology 34(3):468–492.

Hainmueller, Jens, Daniel J. Hopkins and Teppei Yamamoto. 2014. “Causal Inference in

Conjoint Analysis: Understanding Multidimensional Choices via Stated Preference Ex-

periments.” Political Analysis 22(1):1–30.

Holbein, John B. and D. Sunshine Hillygus. 2020. Making Young Voters: Converting Civic

Attitudes into Civic Action. Cambridge University Press.

Hopkins, Daniel J. 2009. “No More Wilder Effect, Never a Whitman Effect: When and Why

Polls Mislead about Black and Female Candidates.” The Journal of Politics 71(3):769–781.

36

Horiuchi, Yusaku. 2005. Institutions, Incentives and Electoral Participation in Japan: Cross-

Level and Cross-National Perspectives. New York: Routledge.

Horiuchi, Yusaku, Daniel M. Smith and Teppei Yamamoto. 2020. “Identifying Voter Pref-

erences for Politicians’ Personal Attributes: A Conjoint Experiment in Japan.” Political

Science Research and Methods 8(1):75–91.

Horiuchi, Yusaku, Tadashi Komatsu and Fumio Nakaya. 2012. “Should Candidates Smile

to Win Elections? An Application of Automated Face Recognition Technology.” Political

Psychology 33(6):925–933.

Inter-Parliamentary Union. 2018. Youth Participation in National Parliaments. Inter-

Parliamentary Union.

Jacobson, Gary C. 1983. The Politics of Congressional Elections. Boston: Little, Brown &

Co.

Joshi, Devin K. 2013. “The Representation of Younger Age Cohorts in Asian Parliaments:

Do Electoral Systems Make a Difference?” Representation 49(1):1–16.

Kage, Rieko, Frances M. Rosenbluth and Seiki Tanaka. 2019. “What Explains Low Fe-

male Political Representation? Evidence From Survey Experiments in Japan.” Politics &

Gender 15(2):285–309.

Kam, Cindy D. and Donald R. Kinder. 2012. “Ethnocentrism as a Short-Term Force in the

2008 American Presidential Election.” American Journal of Political Science 56(2):326–

340.

King, Amy and Andrew Leigh. 2009. “Beautiful Politicians.” Kyklos 62(4):579–593.

Kirkland, Patricia A. and Alexander Coppock. 2017. “Candidate Choice Without Party

Labels: New Insights From Conjoint Survey Experiments.” Political Behavior 40(3):571–

591.

37

Kite, Mary E., Gary D. Stockdale, Bernard E. Witley, Jr. and Blair T. Johnson. 2005. “At-

titudes Toward Younger and Older Adults: An Updated Meta-Analytic Review.” Journal

of Social Issues 61(2):241–266.

Krupnikov, Yanna, Spencer Piston and Nichole M. Bauer. 2015. “Saving Face: Identify-

ing Voter Responses to Black Candidates and Female Candidates.” Political Psychology

37(2):253–273.

Kubeck, Jean E., Norma D. Delp, Tammy K. Haslett and Michael A. McDaniel. 1996. “Does

Job-Related Training Performance Decline With Age?” Psychology and Aging 11(1):92–

107.

Lawless, Jennifer L. and Richard L. Fox. 2015. Running from Office: Why Young Americans

Are Turned Off to Politics. Oxford: Oxford University Press.

Lawson, Chappell, Gabriel S. Lenz, Andy Baker and Michael Myers. 2010. “Looking Like

a Winner: Candidate Appearance and Electoral Success in New Democracies.” World

Politics 62(04):561–593.

Lewis, Jonathan and Brian J Masshardt. 2002. “Election posters in Japan.” Japan Forum

14(3):373–404.

Mansbridge, Jane. 1999. “Should Blacks Represent Blacks and Women Represent Women?

A Contingent ‘Yes’.” The Journal of Politics 61(3):628–657.

McClean, Charles T. 2020. “Does It Matter That Politicians Are Older Than Their Con-

stituents? Yes.” Unpublished manuscript.

McConnaughy, Corrine M., Ismail K. White, David L. Leal and Jason P. Casellas. 2010. “A

Latino on the Ballot: Explaining Coethnic Voting Among Latinos and the Response of

White Americans.” The Journal of Politics 72(4):1199–1211.

38

McElwain, Kenneth Mori. 2008. “Manipulating Electoral Rules to Manufacture Single-Party

Dominance.” American Journal of Political Science 52(1):32–47.

McGoldrick, Ann E. and James Arrowsmith. 2001. Discrimination by Age: The Organi-

zational Response. In Ageism in Work and Employment, ed. Ian Glover and Mohamed

Branine. Burlington, VT: Ashgate pp. 75–96.

Ono, Yoshikuni and Barry C Burden. 2019. “The Contingent Effects of Candidate Sex on

Voter Choice.” Political Behavior 41(3):583–607.

Ono, Yoshikuni and Masahiro Yamada. 2020. “Do Voters Prefer Gender Stereotypic Candi-

dates? Evidence from a Conjoint Survey Experiment in Japan.” Political Science Research

and Methods 8(3):477–492.

Perry, Elissa L., Carol T. Kulik and Anne C. Bourhis. 1996. “Moderating Effects of Personal

and Contextual Factors in Age Discrimination.” Journal of Applied Psychology 81(6):628–

647.

Phelps, Edmund. 1972. “The Statistical Theory of Racism and Sexism.” The American

Economic Review 62(4):659–661.

Phillips, Anne. 1995. The Politics of Presence. Oxford: Oxford University Press.

Pomante, Michael J. and Scot Schraufnagel. 2015. “Candidate Age and Youth Voter

Turnout.” American Politics Research 43(3):479–503.

Posthuma, Richard A. and Michael A. Campion. 2009. “Age Stereotypes in the Workplace:

Common Stereotypes, Moderators, and Future Research Directions.” Journal of Manage-

ment 35(1):158–188.

Rahn, Wendy M. 1993. “The Role of Partisan Stereotypes in Information Processing About

Political Candidates.” American Journal of Political Science 37(2):472–496.

39

Shames, Shauna L. 2017. Out of the Running: Why Millenials Reject Political Careers and

Why It Matters. New York: New York University Press.

Shugart, Matthew SAzberg, Melody Ellis Valdini and Kati Suominen. 2005. “Looking for

Locals: Voter Information Demands and Personal Vote-Earning Attributes of Legislators

under Proportional Representation.” American Journal of Political Science 49(2):437–449.

Smith, Daniel M. and Steven R. Reed. 2018. “The Reed-Smith Japanese House of Repre-

sentatives Elections Dataset.” Harvard Dataverse.

URL: https://doi.org/10.7910/DVN/QFEPXD

Sniderman, Paul M., Richard A. Brody and Philip E. Tetlock. 1991. Reasoning and Choice:

Explorations in Political Psychology. Cambridge: Cambridge University Press.

Stockemer, Daniel and Aksel Sundstrom. 2018. “Age Representation in Parliaments: Can

Institutions Pave the Way for the Young?” European Political Science Review 10(3):467–

490.

Tajfel, Henri and John C. Turner. 1986. The Social Identity Theory of Intergroup Behavior.

In Psychology of Intergroup Relations, ed. S. Worchel and W. G. Austin. Chicago: Nelson-

Hall pp. 7–24.

Teele, Dawn Langan, Joshua Kalla and Frances Rosenbluth. 2018. “The Ties That Double

Bind: Social Roles and Women’s Underrepresentation in Politics.” American Political

Science Review 112(3):525–541.

Terkildsen, Nayda. 1993. “When White Voters Evaluate Black Candidates: The Processing

Implications of Candidate Skin Color, Prejudice, and Self-Monitoring.” American Journal

of Political Science 37(4):1032–1053.

Todorov, Alexander, Anesu N. Mandisodza, Amir Goren and Crystal C. Hall. 2005. “Infer-

40

ences of Competence From Faces Predict Election Outcomes.” Science 308(5728):1623–

1626.

Wattenberg, Martin P. 2007. Is Voting for Young People? New York: Pearson Longman.

Weaver, Vesla M. 2011. “The Electoral Consequences of Skin Color: The ‘Hidden’ Side of

Race in Politics.” Political Behavior 34(1):159–192.

Webster, Steven W. and Andrew W. Pierce. 2019. “Older, Younger, or More Similar? The

Use of Age as a Voting Heuristic.” Social Science Quarterly 100(3):635–652.

41

Appendix

Figure A1: Candidate Age and Vote Choice (Likely Voters)

●

●Older

Younger

(Baseline: Middle−Aged)

−60 −40 −20 0 20Percentage Point Change in Probability of Voter Choosing Candidate

Notes: Dependent variable is equal to 1 if the respondent said they would vote for the candidate, and 0otherwise. Bars show 95% confidence intervals.

42